[Forum SIS] Annuncio seminario A. Farcomeni

[Monia Lupparelli]

ANNUNCIO SEMINARIO
*******************
Giovedì 28 febbraio ore 15.00
Dipartimento di Scienze Statistiche
Via Bella Arti, 41, Bologna
*******************

Alessio FARCOMENI dell'Università La Sapienza di Roma

terrà il seminario dal titolo

"A multivariate generalized linear mixed model based on a latent Markov heterogeneity structure"


Abstract: Latent Markov models (LM) are often used to conveniently model subject-specific time-varying unobserved heterogeneity in longitudinal studies.

There could be anyway additional sources of heterogeneity (e.g., subjects could be clustered in centers, schools, geographical areas) which could lead the basic LM to be biased. The solutions proposed so far use additional (with respect to the latent Markov structure) multivariate distributions for the random effects.

With these approaches in our opinion the interpretability and computational convenience of the LM are lost.

Furthermore, additional unobserved

heterogeneity is time-constant in available formulations.

In this work we propose a (multivariate) LM in which the manifest model is the classical generalized linear mixed model, and all random effects are time-varying. Random effects follow a unique latent Markov structure, thus preserving the LM formulation, interpretability and computational convenience.
We outline the simplest case of perfect dependence in the hidden structure, the most general case of dimension specific time-variability, and describe how to obtain any in-between situation. The model is fit with an EM type algorithm with the same computational complexity of the basic LM model with covariates.


Tutti gli interessati sono invitati a partecipare




Monia Lupparelli, Ph.D.
Dipartimento di Scienze Statistiche
Via Belle Arti, 41
40126 Bologna, IT
Phone +39 051 2098268     Fax +39 051 232153
Home page: http://www2.stat.unibo.it/lupparelli/

-------------- parte successiva --------------
Un allegato HTML è stato rimosso...
URL: <http://www.stat.unipg.it/pipermail/sis/attachments/20130218/94ecf7d4/attachment.html>